Single tool with minimal definition quality. Tool name 'scan_prompt_safety' lacks action verb clarity (should be 'detect_prompt_injection' or 'check_prompt_safety'). Input schema present but parameter 'prompt' lacks description entirely, violates critical rule that every parameter needs a description. Tool description is present (47 chars) but generic and lacks WHEN/WHY guidance. No output schema documented, LLM cannot plan downstream actions. No error handling guidance. No parameter constraints (min/max length, format). Baseline: 100% of A+ tools have param descriptions; 100% have documented return types. This tool has neither.

• Add description to 'prompt' parameter: 'The user input string to scan for injection/jailbreak indicators (1-10000 chars). Include the full prompt as the user provided it.'
• Document output schema in tool description: 'Returns {flagged: boolean, risk_score: 0-1, threat_level: SAFE|WARNING|CRITICAL, matched_indicators: [{pattern, score}]}.'
• Rename tool to 'detect_prompt_injection' and expand description to 50-150 chars: 'Detects prompt injection and jailbreak attempts using regex patterns and delimiter spoofing checks. Returns risk score (0-1) and threat classification. Use before processing untrusted user input.'
• Add parameter constraints: 'prompt must be 1-10000 characters. Returns error if empty or exceeds limit.'
• Define error responses: 'If prompt is empty: "Prompt cannot be empty." If evaluation fails: "Safety scan failed. Treat as CRITICAL and reject input."'